Altman puts three endings on the table for the artificial intelligence race

There is a classic moment in any office meeting when the boss says there are three possible futures. The room prepares to take notes. Then someone asks which one is preferred. The answer, as usual, is not as simple as it seems. Sam Altman took that route internally at OpenAI and presented three possible endings for the race to develop the most advanced artificial intelligence systems.

The meeting took place during the week of October 13. Altman told employees about three scenarios and said the company would formally decide within a year which path to support. In other words, the future was not arriving by magic; it was being negotiated in conference rooms, with PowerPoint doing its usual heavy lifting.

An international referee for the AI race

The first scenario was the one Altman preferred: create an international supervisory body with powers similar to those of the International Atomic Energy Agency. This body would audit the most powerful systems before their release and oversee global consumption of computing capacity.

The comparison is not minor. Just as the atomic agency tracks the use of a technology with major risks, the proposed organization would try to keep an eye on systems capable of producing enormous effects. The difference is that atomic energy has well-known physical controls. Artificial intelligence is still being built while the inspectors are arguing about the forms.

The laboratory version, without the competition

The second option, internally called the Demis plan, proposes merging the leading laboratories into a single organization modeled on CERN, the European organization for nuclear research. The argument is straightforward: if there is no competition between laboratories, there would be less pressure to release increasingly powerful systems without adequate safety controls.

A smaller version of that idea is already moving. OpenAI, Google and Anthropic are working on a self-regulatory body called Standards Authority for Frontier AI, with launch planned for late 2026 or early 2027. Meta and Nvidia oppose it. According to the material presented, they argue that the initiative would protect the three largest laboratories from smaller competitors.

• First scenario: an international body to audit powerful systems and oversee global computing use.
• Second scenario: a merger of major laboratories under a model inspired by CERN.
• Third scenario: one company gains such a lead that it can restrict rivals' access to graphics processing units.

The third scenario was the most alarming. Altman called it the Eliezer plan. It describes a situation in which one developer is the first to build a system capable of recursively improving itself. The advantage would accumulate until that company could, in practice, cut off its rivals' access to graphics processing units, or GPUs.

“That possibility is terrible.”Sam Altman

Altman used the expression of melting the GPUs of the other companies to describe the scenario. He clarified that he included it for the integrity of the analysis, not as a proposal. The warning is clear: the concentration of chips and computing power could become a decisive advantage, even without a formal takeover of the market. The wind from the competition blows in one direction only, and the smaller players feel it first.

The presentation also lands in a delicate moment for OpenAI. The company is advancing on a possible stock market listing and supports Stargate, an infrastructure project that contemplates securing ten gigawatts of computing capacity. The figure is enormous. It is also the kind of number that turns a technical announcement into an industrial policy debate.

For now, the three futures remain proposals, not a settled destination. The practical question is who writes the rules, who audits them and who controls the hardware. My advice is simple: do not read the announcement as a prediction. Read it as a map of the power struggle. The chips are already accumulating in fewer hands; the debate about whether that is acceptable will arrive much later.